add shouldNotTypecheckWith

[soenkehahn]

@CRogers: Awesome library. What do you think about this addition?

[soenkehahn]

The failure on ghc-8 seems to be caused by ghc including the substring as contextual information in the error message. And then of course it'll be in there. Not sure how to address this.

[isovector]

This is exactly the function I've been looking for. Any chance of getting this merged, @CRogers @soenkehahn ?

[CRogers]

If you can get this to pass CI and compile on a recent GHC version, for sure. I just don't have time to do it myself at the moment.

[isovector]

Sounds great. I'll get started.

[soenkehahn]

Glad someone is interested in this. :)

However, IIRC, back then when I experimented with this branch I had problems with tests that should have failed but passed. The issue was roughly this:

• You write a test-case that includes shouldNotTypecheckWith "some error message".
• Your test case has a (unintentional) type error.
• The error message contains -- as part of the expression that fails the typecheck -- the string literal "some error message".
• The test passes, because shouldNotTypecheckWith detects that there was a type error containing the string some error message.

I thought that it was pretty bad that your tests can accidentally pass, because you have an unintentional type error in your testing code. Not sure whether there is a good solution to this.

[isovector]

@soenkehahn I've gotten around this in #15 by just stripping off the context. IMO the primary usecase for this function is to ensure that custom type errors say what you'd like them to, so the context is probably a safe thing to ignore.